Code generation, symbolic math reasoning, and other tasks require LLMs to produce outputs that are both syntactically and semantically correct. Constrained LLM generation is a promising direction to enforce adherence to formal grammar, but prior works have empirically observed that strict enforcement of formal constraints often diminishes the reasoning capabilities of LLMs. In this work, we first provide a theoretical explanation for why constraining LLM outputs to very restrictive grammars that only allow syntactically valid final answers reduces the reasoning capabilities of the model. Second, we demonstrate that by augmenting the output grammar with carefully designed additional rules, it is always possible to preserve the reasoning capabilities of the LLM while ensuring syntactic and semantic correctness in its outputs. Building on these theoretical insights, we propose a reasoning-augmented constrained decoding algorithm, CRANE, which effectively balances the correctness of constrained generation with the flexibility of unconstrained generation. Experiments on multiple open-source LLMs and benchmarks show that CRANE significantly outperforms both state-of-the-art constrained decoding strategies and standard unconstrained decoding, showing up to 10% points accuracy improvement over baselines on challenging symbolic reasoning benchmarks GSM-symbolic and FOLIO.
Vision-language models (VLMs) have achieved remarkable success in scene understanding and perception tasks, enabling robots to plan and execute actions adaptively in dynamic environments. However, most multimodal large language models lack robust 3D scene localization capabilities, limiting their effectiveness in fine-grained robotic operations. Additionally, challenges such as low recognition accuracy, inefficiency, poor transferability, and reliability hinder their use in precision tasks. To address these limitations, we propose a novel framework that integrates a 2D prompt synthesis module by mapping 2D images to point clouds, and incorporates a small language model (SLM) for supervising VLM outputs. The 2D prompt synthesis module enables VLMs, trained on 2D images and text, to autonomously extract precise 3D spatial information without manual intervention, significantly enhancing 3D scene understanding. Meanwhile, the SLM supervises VLM outputs, mitigating hallucinations and ensuring reliable, executable robotic control code generation. Our framework eliminates the need for retraining in new environments, thereby improving cost efficiency and operational robustness. Experimental results that the proposed framework achieved a 96.0\% Task Success Rate (TSR), outperforming other methods. Ablation studies demonstrated the critical role of both the 2D prompt synthesis module and the output supervision module (which, when removed, caused a 67\% TSR drop). These findings validate the framework's effectiveness in improving 3D recognition, task planning, and robotic task execution.
Large Language Models (LLMs) have made significant strides in code generation and problem solving. Current approaches employ external tool-based iterative debuggers that use compiler or other tool-based runtime feedback to refine coarse programs generated by various methods. However, the effectiveness of these approaches heavily relies on the quality of the initial code generation, which remains an open challenge. In this paper, we introduce CodeSim, a novel multi-agent code generation framework that comprehensively addresses the stages of program synthesis-planning, coding, and debugging-through a human-like perception approach. As human verifies their understanding of any algorithms through visual simulation, CodeSim uniquely features a method of plan verification and internal debugging through the step-by-step simulation of input/output. Extensive experiments across seven challenging competitive problem-solving and program synthesis benchmarks demonstrate CodeSim's remarkable code generation capabilities. Our framework achieves new state-of-the-art (pass@1) results-(HumanEval 95.1%, MBPP 90.7%, APPS 22%, and CodeContests 29.1%). Furthermore, our method shows potential for even greater enhancement when cascaded with external debuggers. To facilitate further research and development in this area, we have open-sourced our framework in this link (https://kagnlp.github.io/codesim.github.io/).
Recent progress in large language models (LLMs) highlights the power of scaling test-time compute to achieve strong performance on complex tasks, such as mathematical reasoning and code generation. This raises a critical question: how should model training be modified to optimize performance under a subsequent test-time compute strategy and budget? To explore this, we focus on pass@N, a simple test-time strategy that searches for a correct answer in $N$ independent samples. We show, surprisingly, that training with cross-entropy (CE) loss can be ${\it misaligned}$ with pass@N in that pass@N accuracy ${\it decreases}$ with longer training. We explain the origins of this misalignment in terms of model overconfidence induced by CE, and experimentally verify our prediction of overconfidence as an impediment to scaling test-time compute via pass@N. Furthermore we suggest a principled, modified training loss that is better aligned to pass@N by limiting model confidence and rescuing pass@N test performance. Our algorithm demonstrates improved mathematical reasoning on MATH and MiniF2F benchmarks under several scenarios: (1) providing answers to math questions; and (2) proving theorems by searching over proof trees of varying shapes. Overall our work underscores the importance of co-designing two traditionally separate phases of LLM development: training-time protocols and test-time search and reasoning strategies.
Previous multilingual benchmarks focus primarily on simple understanding tasks, but for large language models(LLMs), we emphasize proficiency in instruction following, reasoning, long context understanding, code generation, and so on. However, measuring these advanced capabilities across languages is underexplored. To address the disparity, we introduce BenchMAX, a multi-way multilingual evaluation benchmark that allows for fair comparisons of these important abilities across languages. To maintain high quality, three distinct native-speaking annotators independently annotate each sample within all tasks after the data was machine-translated from English into 16 other languages. Additionally, we present a novel translation challenge stemming from dataset construction. Extensive experiments on BenchMAX reveal varying effectiveness of core capabilities across languages, highlighting performance gaps that cannot be bridged by simply scaling up model size. BenchMAX serves as a comprehensive multilingual evaluation platform, providing a promising test bed to promote the development of multilingual language models. The dataset and code are publicly accessible.
Reasoning is a fundamental capability of Large Language Models. While prior research predominantly focuses on enhancing narrow skills like math or code generation, improving performance on many other reasoning tasks remains challenging due to sparse and fragmented training data. To address this issue, we propose CodeI/O, a novel approach that systematically condenses diverse reasoning patterns inherently embedded in contextually-grounded codes, through transforming the original code into a code input-output prediction format. By training models to predict inputs/outputs given code and test cases entirely in natural language as Chain-of-Thought (CoT) rationales, we expose them to universal reasoning primitives -- like logic flow planning, state-space searching, decision tree traversal, and modular decomposition -- while decoupling structured reasoning from code-specific syntax and preserving procedural rigor. Experimental results demonstrate CodeI/O leads to consistent improvements across symbolic, scientific, logic, math & numerical, and commonsense reasoning tasks. By matching the existing ground-truth outputs or re-executing the code with predicted inputs, we can verify each prediction and further enhance the CoTs through multi-turn revision, resulting in CodeI/O++ and achieving higher performance. Our data and models are available at https://github.com/hkust-nlp/CodeIO.
Large language models (LLMs) have demonstrated remarkable code generation capabilities, but the correctness of the generated code cannot be inherently trusted. This paper explores the feasibility of using formal software verification, specifically the SPARK framework for Ada, to ensure the reliability of LLM-generated code. We present Marmaragan, a tool that leverages an LLM in order to generate SPARK annotations for existing programs, enabling formal verification of the code. The tool is benchmarked on a curated set of SPARK programs, with annotations selectively removed to test specific capabilities. The performance of Marmaragan with GPT-4o on the benchmark is promising, with correct annotations having been generated for 50.7% of the benchmark cases. The results establish a foundation for future work on combining the power of LLMs with the reliability of formal software verification.
The prototyping of computer games, particularly card games, requires extensive human effort in creative ideation and gameplay evaluation. Recent advances in Large Language Models (LLMs) offer opportunities to automate and streamline these processes. However, it remains challenging for LLMs to design novel game mechanics beyond existing databases, generate consistent gameplay environments, and develop scalable gameplay AI for large-scale evaluations. This paper addresses these challenges by introducing a comprehensive automated card game prototyping framework. The approach highlights a graph-based indexing method for generating novel game designs, an LLM-driven system for consistent game code generation validated by gameplay records, and a gameplay AI constructing method that uses an ensemble of LLM-generated action-value functions optimized through self-play. These contributions aim to accelerate card game prototyping, reduce human labor, and lower barriers to entry for game developers.
Language Models (LLMs), such as transformer-based neural networks trained on billions of parameters, have become increasingly prevalent in software engineering (SE). These models, trained on extensive datasets that include code repositories, exhibit remarkable capabilities for SE tasks. However, evaluating their effectiveness poses significant challenges, primarily due to the potential overlap between the datasets used for training and those employed for evaluation. To address this issue, we introduce SnipGen, a comprehensive repository mining framework designed to leverage prompt engineering across various downstream tasks for code generation. SnipGen aims to mitigate data contamination by generating robust testbeds and crafting tailored data points to assist researchers and practitioners in evaluating LLMs for code-related tasks. In our exploratory study, SnipGen mined approximately 227K data points from 338K recent code changes in GitHub commits, focusing on method-level granularity. SnipGen features a collection of prompt templates that can be combined to create a Chain-of-Thought-like sequence of prompts, enabling a nuanced assessment of LLMs' code generation quality. By providing the mining tool, the methodology, and the dataset, SnipGen empowers researchers and practitioners to rigorously evaluate and interpret LLMs' performance in software engineering contexts.
Large Language Models (LLMs) are widely utilized in software engineering (SE) tasks, such as code generation and automated program repair. However, their reliance on extensive and often undisclosed pre-training datasets raises significant concerns about data leakage, where the evaluation benchmark data is unintentionally ``seen'' by LLMs during the model's construction phase. The data leakage issue could largely undermine the validity of LLM-based research and evaluations. Despite the increasing use of LLMs in the SE community, there is no comprehensive study that assesses the extent of data leakage in SE benchmarks for LLMs yet. To address this gap, this paper presents the first large-scale analysis of data leakage in 83 SE benchmarks concerning LLMs. Our results show that in general, data leakage in SE benchmarks is minimal, with average leakage ratios of only 4.8\%, 2.8\%, and 0.7\% for Python, Java, and C/C++ benchmarks, respectively. However, some benchmarks exhibit relatively higher leakage ratios, which raises concerns about their bias in evaluation. For instance, QuixBugs and BigCloneBench have leakage ratios of 100.0\% and 55.7\%, respectively. Furthermore, we observe that data leakage has a substantial impact on LLM evaluation. We also identify key causes of high data leakage, such as the direct inclusion of benchmark data in pre-training datasets and the use of coding platforms like LeetCode for benchmark construction. To address the data leakage, we introduce \textbf{LessLeak-Bench}, a new benchmark that removes leaked samples from the 83 SE benchmarks, enabling more reliable LLM evaluations in future research. Our study enhances the understanding of data leakage in SE benchmarks and provides valuable insights for future research involving LLMs in SE.
Recently, large language models (LLMs) have been deployed to tackle various software engineering (SE) tasks like code generation, significantly advancing the automation of SE tasks. However, assessing the quality of these LLM-generated code and text remains challenging. The commonly used Pass@k metric necessitates extensive unit tests and configured environments, demands a high labor cost, and is not suitable for evaluating LLM-generated text. Conventional metrics like BLEU, which measure only lexical rather than semantic similarity, have also come under scrutiny. In response, a new trend has emerged to employ LLMs for automated evaluation, known as LLM-as-a-judge. These LLM-as-a-judge methods are claimed to better mimic human assessment than conventional metrics without relying on high-quality reference answers. Nevertheless, their exact human alignment in SE tasks remains unexplored. In this paper, we empirically explore LLM-as-a-judge methods for evaluating SE tasks, focusing on their alignment with human judgments. We select seven LLM-as-a-judge methods that utilize general-purpose LLMs, alongside two LLMs specifically fine-tuned for evaluation. After generating and manually scoring LLM responses on three recent SE datasets of code translation, code generation, and code summarization, we then prompt these methods to evaluate each response. Finally, we compare the scores generated by these methods with human evaluation. The results indicate that output-based methods reach the highest Pearson correlation of 81.32 and 68.51 with human scores in code translation and generation, achieving near-human evaluation, noticeably outperforming ChrF++, one of the best conventional metrics, at 34.23 and 64.92. Such output-based methods prompt LLMs to output judgments directly, and exhibit more balanced score distributions that resemble human score patterns. Finally, we provide...
We introduce the Formally Verified Automated Programming Progress Standards, or FVAPPS, a benchmark of 4715 samples for writing programs and proving their correctness, the largest formal verification benchmark, including 1083 curated and quality controlled samples. Previously, APPS provided a benchmark and dataset for programming puzzles to be completed in Python and checked against unit tests, of the kind seen in technical assessments in the software engineering industry. Building upon recent approaches for benchmarks in interactive theorem proving, we generalize the unit tests to Lean 4 theorems given without proof (i.e., using Lean's "sorry" keyword). On the 406 theorems of 100 randomly selected samples, Sonnet correctly proves 30% and Gemini correctly proves 18%. We challenge the machine learning and program synthesis communities to solve both each general purpose programming problem and its associated correctness specifications. The benchmark is available at https://huggingface.co/datasets/quinn-dougherty/fvapps.
Prompt engineering reduces reasoning mistakes in Large Language Models (LLMs). However, its effectiveness in mitigating vulnerabilities in LLM-generated code remains underexplored. To address this gap, we implemented a benchmark to automatically assess the impact of various prompt engineering strategies on code security. Our benchmark leverages two peer-reviewed prompt datasets and employs static scanners to evaluate code security at scale. We tested multiple prompt engineering techniques on GPT-3.5-turbo, GPT-4o, and GPT-4o-mini. Our results show that for GPT-4o and GPT-4o-mini, a security-focused prompt prefix can reduce the occurrence of security vulnerabilities by up to 56%. Additionally, all tested models demonstrated the ability to detect and repair between 41.9% and 68.7% of vulnerabilities in previously generated code when using iterative prompting techniques. Finally, we introduce a "prompt agent" that demonstrates how the most effective techniques can be applied in real-world development workflows.
Large Language Models for Code (LLMs4Code) excel at code generation tasks, yielding promise to release developers from huge software development burdens. Nonetheless, these models have been shown to suffer from the significant privacy risks due to the potential leakage of sensitive information embedded during training, known as the memorization problem. Addressing this issue is crucial for ensuring privacy compliance and upholding user trust, but till now there is a dearth of dedicated studies in the literature that focus on this specific direction. Recently, machine unlearning has emerged as a promising solution by enabling models to "forget" sensitive information without full retraining, offering an efficient and scalable approach compared to traditional data cleaning methods. In this paper, we empirically evaluate the effectiveness of unlearning techniques for addressing privacy concerns in LLMs4Code.Specifically, we investigate three state-of-the-art unlearning algorithms and three well-known open-sourced LLMs4Code, on a benchmark that takes into consideration both the privacy data to be forgotten as well as the code generation capabilites of these models. Results show that it is feasible to mitigate the privacy concerns of LLMs4Code through machine unlearning while maintain their code generation capabilities at the same time. We also dissect the forms of privacy protection/leakage after unlearning and observe that there is a shift from direct leakage to indirect leakage, which underscores the need for future studies addressing this risk.
In this paper, we propose CodeSCM, a Structural Causal Model (SCM) for analyzing multi-modal code generation using large language models (LLMs). By applying interventions to CodeSCM, we measure the causal effects of different prompt modalities, such as natural language, code, and input-output examples, on the model. CodeSCM introduces latent mediator variables to separate the code and natural language semantics of a multi-modal code generation prompt. Using the principles of Causal Mediation Analysis on these mediators we quantify direct effects representing the model's spurious leanings. We find that, in addition to natural language instructions, input-output examples significantly influence code generation.
Natural Language to Visualization (NL2Vis) seeks to convert natural-language descriptions into visual representations of given tables, empowering users to derive insights from large-scale data. Recent advancements in Large Language Models (LLMs) show promise in automating code generation to transform tabular data into accessible visualizations. However, they often struggle with complex queries that require reasoning across multiple tables. To address this limitation, we propose a collaborative agent workflow, termed nvAgent, for NL2Vis. Specifically, nvAgent comprises three agents: a processor agent for database processing and context filtering, a composer agent for planning visualization generation, and a validator agent for code translation and output verification. Comprehensive evaluations on the new VisEval benchmark demonstrate that nvAgent consistently surpasses state-of-the-art baselines, achieving a 7.88% improvement in single-table and a 9.23% improvement in multi-table scenarios. Qualitative analyses further highlight that nvAgent maintains nearly a 20% performance margin over previous models, underscoring its capacity to produce high-quality visual representations from complex, heterogeneous data sources.
Scientific data visualization is pivotal for transforming raw data into comprehensible visual representations, enabling pattern recognition, forecasting, and the presentation of data-driven insights. However, novice users often face difficulties due to the complexity of selecting appropriate tools and mastering visualization techniques. Large Language Models (LLMs) have recently demonstrated potential in assisting code generation, though they struggle with accuracy and require iterative debugging. In this paper, we propose PlotGen, a novel multi-agent framework aimed at automating the creation of precise scientific visualizations. PlotGen orchestrates multiple LLM-based agents, including a Query Planning Agent that breaks down complex user requests into executable steps, a Code Generation Agent that converts pseudocode into executable Python code, and three retrieval feedback agents - a Numeric Feedback Agent, a Lexical Feedback Agent, and a Visual Feedback Agent - that leverage multimodal LLMs to iteratively refine the data accuracy, textual labels, and visual correctness of generated plots via self-reflection. Extensive experiments show that PlotGen outperforms strong baselines, achieving a 4-6 percent improvement on the MatPlotBench dataset, leading to enhanced user trust in LLM-generated visualizations and improved novice productivity due to a reduction in debugging time needed for plot errors.
Despite the fact that large language models (LLMs) show exceptional skill in instruction following tasks, this strength can turn into a vulnerability when the models are required to disregard certain instructions. Instruction-following tasks typically involve a clear task description and input text containing the target data to be processed. However, when the input itself resembles an instruction, confusion may arise, even if there is explicit prompting to distinguish between the task instruction and the input. We refer to this phenomenon as instructional distraction. In this paper, we introduce a novel benchmark, named DIM-Bench, specifically designed to assess LLMs' performance under instructional distraction. The benchmark categorizes real-world instances of instructional distraction and evaluates LLMs across four instruction tasks: rewriting, proofreading, translation, and style transfer -- alongside five input tasks: reasoning, code generation, mathematical reasoning, bias detection, and question answering. Our experimental results reveal that even the most advanced LLMs are susceptible to instructional distraction, often failing to accurately follow user intent in such cases.
Black-box algorithms are designed to optimize functions without relying on their underlying analytical structure or gradient information, making them essential when gradients are inaccessible or difficult to compute. Traditional methods for solving black-box optimization (BBO) problems predominantly rely on non-parametric models and struggle to scale to large input spaces. Conversely, parametric methods that model the function with neural estimators and obtain gradient signals via backpropagation may suffer from significant gradient errors. A recent alternative, Explicit Gradient Learning (EGL), which directly learns the gradient using a first-order Taylor approximation, has demonstrated superior performance over both parametric and non-parametric methods. In this work, we propose two novel gradient learning variants to address the robustness challenges posed by high-dimensional, complex, and highly non-linear problems. Optimistic Gradient Learning (OGL) introduces a bias toward lower regions in the function landscape, while Higher-order Gradient Learning (HGL) incorporates second-order Taylor corrections to improve gradient accuracy. We combine these approaches into the unified OHGL algorithm, achieving state-of-the-art (SOTA) performance on the synthetic COCO suite. Additionally, we demonstrate OHGLs applicability to high-dimensional real-world machine learning (ML) tasks such as adversarial training and code generation. Our results highlight OHGLs ability to generate stronger candidates, offering a valuable tool for ML researchers and practitioners tackling high-dimensional, non-linear optimization challenges
Large Language Models (LLMs) have made substantial strides in structured tasks through Reinforcement Learning (RL), demonstrating proficiency in mathematical reasoning and code generation. However, applying RL in broader domains like chatbots and content generation -- through the process known as Reinforcement Learning from Human Feedback (RLHF) -- presents unique challenges. Reward models in RLHF are critical, acting as proxies that evaluate the alignment of LLM outputs with human intent. Despite advancements, the development of reward models is hindered by challenges such as computational heavy training, costly evaluation, and therefore poor reproducibility. We advocate for using embedding-based input in reward model research as an accelerated solution to those challenges. By leveraging embeddings for reward modeling, we can enhance reproducibility, reduce computational demands on hardware, improve training stability, and significantly reduce training and evaluation costs, hence facilitating fair and efficient comparisons in this active research area. We then show a case study of reproducing existing reward model ensemble research using embedding-based reward models. We discussed future avenues for research, aiming to contribute to safer and more effective LLM deployments.